>This should be fixed by the changes I made recently in
choose_bitmap_and
>--- it wasn't being aggressive about pruning overlapping AND conditions
when a sub-OR was involved. It's possible the new coding is >*too*
aggressive, and will reject indexes that it'd be profitable to include;
but at least it won't make this particular mistake.
Ok, cool. I don't have time to test this right now as the project has to
move on (and I guess testing the fix would require a dump+build CVS
version+restore), but as a temporary workaround I simly dropped the
xda_dat index (all queries on that table include the
person_information__id column anyway).
- Mikael
